I. Introduction

Like all the sharing systems we know, Airbnb, Uber etc., the Citi Bike is also a sharing system which is the network of bicycle rental stations intended for point-to-point transportation. The Citi Bike is also the largest bike sharing system in NYC, with 12000 bikes and 750 stations across Manhattan, Brooklyn and Queens. It’s a fast, healthy, sustainable and highly popular way to get around NYC.

People can rent bikes at various docking stations throughout the city and return them to another docking station. There are two main forms of payment: The first one is “pay as you go” means you pay for per ride. The second one is “Annual Subscription” means you pay a flat fee for the year with unlimited rides and higher cap on the ride. There is also a time limit on how long the bike can be in use per ride; 30 minutes for non-subscribers and 45 minutes for subscribers. Financial penalties are applied in the cases the ride exceed these limits.

In our project, we will mainly focus on two parts about Citi Bike Dataset. For the first part, we will focus on the general analysis about this dataset, including the analysis of Age, Gender, Usertype, the most popular routes or stations etc. In addition to analysis about the data, we will come up some recommendations based on the findings which we explore from the plots. For the second part, we will focus on and interactive visualization. Our shiny app shows more details about some of the questions we analyzed on the first part, displaying more geographical information and providing new insights.

II. Description of the data source

We downloaded the data from Citi Bike’s official website: https://www.citibikenyc.com/system-data

We will use the data from January 2018 to December 2018, and there are 17548339 records in total.

From the raw data, there are 15 variables :

Trip Duration (seconds) : Length of ride in seconds

Start Time and Date : Date and Time of starting of the ride

Stop Time and Date : Date and Time of ending of the ride

Start Station Name : Station name of starting ride

End Station Name : Station name of ending ride

Start Station ID : Station ID of starting ride

End Station ID : Station ID of ending ride

Start Station Latitude : Latitude of starting station

Start Station Longitude : Longitude of starting station

End Station Latitude : Latitude of ending station

End Station Longitude : Longitude of ending station

Bike ID

User Type (Customer = 24-hour pass or 3-day pass user; Subscriber = Annual Member)

Gender (Zero=unknown; 1=male; 2=female)

Year of Birth

# import data
CitiBike<- data.frame()
ListFiles <- paste("2018", c("01","02","03", "04", "05", "06", "07", "08","09","10","11","12"), ".csv", sep = "")
for(file in ListFiles) {
  CitiBike <- rbind(CitiBike, read.csv(file, header = T, stringsAsFactors = F))
}

# select columns that we use
CitiBike <- CitiBike[,-c(4,6,7,8,10,11,12)]
saveRDS(CitiBike, "trip.rds")

# load data from rds file
CitiBike <- readRDS("trip.rds")

III. Description of data import / cleaning / transformation

Firstly, we did the data cleaning in R. But we found there are no missing values in our dataset.

# check missing values
CitiBike[!complete.cases(CitiBike),]

## [1] tripduration       starttime          stoptime          
## [4] start.station.name end.station.name   usertype          
## [7] birth.year         gender            
## <0 rows> (or 0-length row.names)

Then, we did some data transformations.

1.Create two new variables in our dataset, “StartDate” and “StartTime”, extracting the exact date and time from “Start Time and Date” variable.

# date and time extract
CitiBike$StartDate <- as.Date(factor(CitiBike$starttime),format="%Y-%m-%d")
CitiBike$StartTime <- as.POSIXct(strptime(CitiBike$starttime, "%Y-%m-%d %H:%M:%S"))
CitiBike$StartTime <- strftime(CitiBike$StartTime, format ="%H:%M:%S")

2.Create a new variable “wday” which stores the Weekdays for the start date.

3.Create another new variable named “Minutes”, which converts the trip duration time from seconds to minutes.

4.Create an “Age” variable, equal to 2018 minus “birth.year”.

#Weekdays and Weekend for Start Date
CitiBike$wday <- wday(CitiBike$StartDate, label = TRUE, abbr = FALSE)

#Trip Minutes
CitiBike$Minutes <- CitiBike$tripduration/60

#Age
CitiBike$Age <- 2018 - CitiBike$birth.year

5.Convert the “gender” variable from numbers to name. (0=Unknown; 1=Male; 2=Female)

# convert gender
CitiBike$gender <- ifelse(CitiBike$gender==1,"Male",ifelse(CitiBike$gender==2,"Female","Unknown"))

6.Create “Month” variable from StartDate.

# get month
CitiBike$Month <- as.numeric(substr(CitiBike$starttime,6,7))

Then, we check for outliers (values that might not be reasonable).

Drop the extreme values for Trip Duration Time (Minutes)

We define a function F(t) which helps us calculate the quantile of duration time. From this function, we can calculate the result that 99.9% of duration time is within 265 minutes. So we remove all the records with more than 265 minutes in duration time. In this step, we remove 17485 records in total.

# check quantiles
Ft<-ecdf(CitiBike[order(CitiBike$Minutes),12])
q=quantile(Ft,c(0.85,0.95,0.99,0.999))

Drop the outliers for age

After calculating the riders’ age from their birth year, we found there are some outliers. So we remove all the records with more than 100 years old. In this step, we remove 11141 records in total.

# remove outliers in minutes and age
CitiBike<-CitiBike[CitiBike$Minutes<=265,]
CitiBike <- CitiBike[CitiBike$Age<=100,]

IV. Analysis of missing values

Our dataset has no missing values. Because the data has been pre-cleaned before uploading. From Citi Bike’s introduction to dataset on their official website, they have already removed trips that are taken by staff as they service and inspect the system, trips that are taken to/from any of their “test” stations (which they were using more in June and July 2013), and any trips that were below 60 seconds in length (potentially false starts or users trying to re-dock a bike to ensure it’s secure).

V. Results

The following are some exploratory data analysis that we have explored.

How many trips on each day throughout the year?

UsagePerDay <- CitiBike %>%
                group_by(StartDate) %>%
                tally()

ggplot(data=UsagePerDay, aes(x=StartDate, y=n)) +
  geom_area(fill="lightblue",color="blue") +
  ylab("Number of trips") +
  xlab("2018") +
  ggtitle("NYC Daily Citi Bike Trips") +
  theme(plot.title = element_text(hjust=0.5))

The plot shows a general trend of increasing number of trips from Winter to Summer, and decreasing number of trips from Summer to Winter. And we can also observe there are sharp decrease daily trip counts, corresponding to weekends and holidays. As there are more trips around summer, starting from May to Oct, we would recommend distribute more bikes during this period. After seeing the general trend for daily trips, we shall have a look at the monthly plot throughout the year.

How many trips on each Month throughout the year?

UsagePerMonth<-CitiBike %>%
                  group_by(Month) %>%
                  tally()
ggplot(data=UsagePerMonth, aes(x=Month, y=n, group = 1, label=n)) +
  ylim(500000,2100000)+
  geom_point() +
  geom_line() +
  geom_text(vjust=-1, size = 3)+
  ylab("Number of trips") +
  xlab("") +
  ggtitle("NYC Monthly Citi Bike Trips") +
  theme(plot.title = element_text(hjust=0.5))

For the monthly plot, we use the line chart to represent the trend of total trips for each month. Not surprisingly, there are dramatically fewer trips during the cold winter and more trips during the hot summer, which has the same trend with daily plot. In August, the usage reaches its peak, the number of trips in August is 1974296. Actually from May to October, the usage almost are the same, but there is a huge decrease from October to November and a dramatic increase from April to May. As we mentioned before, because there are more trips starting from May to Oct, we would recommend distributing more bikes during this period.

Distribution of trip duration

ggplot(data = CitiBike, aes(x=Minutes)) +
  geom_bar(fill="steelblue") +
  xlim(0,60) +
  ylab("Number of trips") +
  ggtitle("Distribution of Trip Duration") +
  theme(plot.title = element_text(hjust=0.5))

This plot tells us that for those who finished trip in 60 minutes, most of them finished their trip very quickly, in the range of from 2 to 20 minutes. We guess most of them are one-time customers and just trying to take a ride to have fun. Maybe they’re easier to get bored and tired, then finish the trip very soon. Based on this, we would recommend to increase the price for the first 20 minutes and lower the price after 20 minutes.